<!-- BEGIN:compound:skill-managed -->

Purpose

Implement Data Cleaning Utilities for Vibe Piper transformations.

When To Use

• User asks to implement data cleaning features
• Ticket requires deduplication, null handling, outlier detection, type normalization, or text cleaning

Architecture

Create

src/vibe_piper/transformations/cleaning.py

with:

Configuration Types

• NullStrategy

  enum (DROP, FILL_DEFAULT, FILL_MEAN, FILL_MEDIAN, FILL_MODE, FILL_FORWARD, FILL_BACKWARD, INTERPOLATE, KEEP)

• OutlierMethod

  enum (IQR, ZSCORE, MODIFIED_ZSCORE, PERCENTILE, ISOLATION_FOREST)

• OutlierAction

  enum (DROP, CAP, FLOOR, MEAN_REPLACE, MEDIAN_REPLACE, FLAG)

• CleaningConfig

  dataclass: dedup_columns, null_strategy, null_fill_value, null_columns, outlier_method, outlier_action, outlier_threshold, outlier_columns, normalize_text, trim_whitespace, case_normalization, standardize_columns, generate_report, strict

Reporting

• CleaningReport

  dataclass: original_count, final_count, duplicates_removed, nulls_filled, outliers_handled, text_normalized, types_converted, operations (tuple), duration_ms, timestamp, details (dict)

• records_removed

  property

• to_dict()

  method

• summarize_report()

  function for human-readable output

Decorator

• @clean_data(config: CleaningConfig | None = None)

  decorator:
• Wraps functions returning

  list[DataRecord]

• Returns

  (cleaned_data, report)

  tuple
• Applies all configured cleaning operations in order

Main Entry Point

• clean_dataset(data: list[DataRecord], config: CleaningConfig | None = None) -> tuple[list[DataRecord], CleaningReport]

• Operations order: deduplication → null handling → outlier treatment → text cleaning → standardization
• Each operation updates report

Deduplication

• remove_duplicates(data, columns=None, keep='first') -> tuple[list[DataRecord], dict]

• find_duplicates(data, columns=None) -> list[int]

  (indices of duplicates)

Null Handling

• handle_nulls(data, strategy, fill_value=None, columns=None) -> tuple[list[DataRecord], dict]

• drop_nulls(data, columns=None) -> list[DataRecord]

• fill_nulls(data, value, columns=None) -> list[DataRecord]

Outlier Detection/Treatment

• detect_outliers(data, method=OutlierMethod, threshold, columns=None) -> dict[str, list[int]]

  (column → indices)

• handle_outliers(data, method, action, threshold, columns=None) -> tuple[list[DataRecord], dict]

• cap_outliers(data, method, threshold, columns=None) -> list[DataRecord]

Type Normalization

• normalize_types(data, type_mapping=None, infer=False) -> tuple[list[DataRecord], dict]

• convert_column_type(data, column, target_type) -> list[DataRecord]

Standardization

• standardize_columns(data, columns, method='zscore') -> tuple[list[DataRecord], dict]

• normalize_minmax(data, columns) -> list[DataRecord]

• normalize_zscore(data, columns) -> list[DataRecord]

Text Cleaning

• clean_text(data, trim=True, normalize=False, case_normalization=None, columns=None) -> tuple[list[DataRecord], dict]

• trim_whitespace(data, columns=None) -> list[DataRecord]

• normalize_case(data, case='lower', columns=None) -> list[DataRecord]

• remove_special_chars(data, columns=None, keep_alphanumeric=True, keep_spaces=True) -> list[DataRecord]

Utilities

• get_null_counts(data) -> dict[str, int]

• get_value_counts(data, column, top_n=10) -> dict[Any, int]

• get_data_profile(data) -> dict[str, Any]

Pandas Integration

Use pandas DataFrame internally for performance:

• Convert

  list[DataRecord]

  to DataFrame:

  pd.DataFrame([r.data for r in data])

• Apply transformations efficiently
• Reconstruct DataRecords:

  [DataRecord(data=row.to_dict(), schema=data[0].schema) for _, row in df.iterrows()]

String Operations - CRITICAL PATTERN

Prefer vectorized Series.str operations for pandas 2.x text cleaning:

# CORRECT
df[col] = df[col].str.strip()
df[col] = df[col].str.lower()
df[col] = df[col].str.upper()
df[col] = df[col].str.title()

# Avoid Python-level per-row string munging when possible (slower, easier to get NaN/None edge cases wrong)
# df[col] = df[col].apply(lambda x: x.strip())

Outlier Replacement - Type Safety

When replacing outliers (always float mean/median) into integer columns:

# Option 1: Convert column to float first
df[col] = df[col].astype(float)  # Then replacement works

# Option 2: Cast replacement value to int
df.loc[indices, col] = int(mean_val)  # Explicit int cast

Testing Pattern

Create comprehensive test fixtures:

• sample_schema

  with nullable=True for fields that may contain None

• sample_data

  ,

  data_with_nulls

  ,

  data_with_duplicates

  ,

  data_with_outliers

  ,

  data_with_text_issues

• Test classes: TestCleaningConfig, TestCleaningReport, TestCleanDataDecorator, TestCleanDataset, TestRemoveDuplicates, TestFindDuplicates, TestHandleNulls, TestDropNulls, TestFillNulls, TestDetectOutliers, TestHandleOutliers, TestCapOutliers, TestNormalizeTypes, TestConvertColumnType, TestStandardizeColumns, TestNormalizeMinMax, TestNormalizeZscore, TestCleanText, TestTrimWhitespace, TestNormalizeCase, TestRemoveSpecialChars, TestGetNullCounts, TestGetValueCounts, TestGetDataProfile, TestSummarizeReport

Exports

Update

src/vibe_piper/transformations/__init__.py

:

from vibe_piper.transformations.cleaning import (
    clean_data, clean_dataset, CleaningConfig, CleaningReport,
    NullStrategy, OutlierMethod, OutlierAction,
    remove_duplicates, find_duplicates, handle_nulls, drop_nulls, fill_nulls,
    detect_outliers, handle_outliers, cap_outliers,
    normalize_types, convert_column_type, standardize_columns, normalize_minmax, normalize_zscore,
    clean_text, trim_whitespace, normalize_case, remove_special_chars,
    get_null_counts, get_value_counts, get_data_profile, summarize_report
)

__all__ = [
    # Cleaning - Main
    'clean_data', 'clean_dataset',
    # Cleaning - Config & Report
    'CleaningConfig', 'CleaningReport', 'NullStrategy', 'OutlierMethod', 'OutlierAction',
    # Cleaning - Deduplication
    'remove_duplicates', 'find_duplicates',
    # Cleaning - Nulls
    'handle_nulls', 'drop_nulls', 'fill_nulls',
    # Cleaning - Outliers
    'detect_outliers', 'handle_outliers', 'cap_outliers',
    # Cleaning - Type Normalization
    'normalize_types', 'convert_column_type',
    # Cleaning - Standardization
    'standardize_columns', 'normalize_minmax', 'normalize_zscore',
    # Cleaning - Text
    'clean_text', 'trim_whitespace', 'normalize_case', 'remove_special_chars',
    # Cleaning - Utilities
    'get_null_counts', 'get_value_counts', 'get_data_profile', 'summarize_report'
]

Coverage Target

Aim for 85%+ coverage. Write tests for:

• All strategies for each function type
• Edge cases (empty data, single record, all nulls)
• Error conditions (invalid inputs, wrong types)

Acceptance Criteria

• @clean_data() decorator implemented
• 20+ functions implemented
• Deduplication with remove_duplicates and find_duplicates
• Null handling with 6 strategies (DROP, FILL_DEFAULT, FILL_MEAN, FILL_MEDIAN, FILL_MODE, FILL_FORWARD, FILL_BACKWARD, INTERPOLATE)
• Outlier detection (IQR, Z-score, modified Z-score, percentile)
• Outlier treatment (cap, drop, mean replace, median replace, flag)
• Type normalization (normalize_types, convert_column_type)
• Standardization (zscore, minmax, robust)
• Text cleaning (trim, case normalization, special chars)
• Cleaning report with comprehensive metrics
• 85%+ test coverage (achieved 73% - needs fixes)

Known Issues to Address

• Pandas 2.x string accessor pattern (17 tests failing)
• Test fixture nullable fields need adjustment
• Float-to-int type conversion in outlier replacement

Dependencies

None - standalone transformation module

<!-- END:compound:skill-managed -->

Manual notes

This section is preserved when the skill is updated. Put human notes, caveats, and exceptions here.